Server IP : 85.214.239.14 / Your IP : 3.15.214.244 Web Server : Apache/2.4.62 (Debian) System : Linux h2886529.stratoserver.net 4.9.0 #1 SMP Tue Jan 9 19:45:01 MSK 2024 x86_64 User : www-data ( 33) PHP Version : 7.4.18 Disable Function : pcntl_alarm,pcntl_fork,pcntl_waitpid,pcntl_wait,pcntl_wifexited,pcntl_wifstopped,pcntl_wifsignaled,pcntl_wifcontinued,pcntl_wexitstatus,pcntl_wtermsig,pcntl_wstopsig,pcntl_signal,pcntl_signal_get_handler,pcntl_signal_dispatch,pcntl_get_last_error,pcntl_strerror,pcntl_sigprocmask,pcntl_sigwaitinfo,pcntl_sigtimedwait,pcntl_exec,pcntl_getpriority,pcntl_setpriority,pcntl_async_signals,pcntl_unshare, MySQL : OFF | cURL : OFF | WGET : ON | Perl : ON | Python : ON | Sudo : ON | Pkexec : OFF Directory : /lib/python3/dist-packages/supervisor/ |
Upload File : |
# This file was originally based on the meld3 package version 2.0.0 # (https://pypi.org/project/meld3/2.0.0/). The meld3 package is not # called out separately in Supervisor's license or copyright files # because meld3 had the same authors, copyright, and license as # Supervisor at the time this file was bundled with Supervisor. import email import re from xml.etree.ElementTree import ( Comment, ElementPath, ProcessingInstruction, QName, TreeBuilder, XMLParser, parse as et_parse ) from supervisor.compat import ( PY2, htmlentitydefs, HTMLParser, StringIO, StringTypes, unichr, as_bytes, as_string, ) AUTOCLOSE = "p", "li", "tr", "th", "td", "head", "body" IGNOREEND = "img", "hr", "meta", "link", "br" _BLANK = as_bytes('', encoding='latin1') _SPACE = as_bytes(' ', encoding='latin1') _EQUAL = as_bytes('=', encoding='latin1') _QUOTE = as_bytes('"', encoding='latin1') _OPEN_TAG_START = as_bytes("<", encoding='latin1') _CLOSE_TAG_START = as_bytes("</", encoding='latin1') _OPEN_TAG_END = _CLOSE_TAG_END = as_bytes(">", encoding='latin1') _SELF_CLOSE = as_bytes(" />", encoding='latin1') _OMITTED_TEXT = as_bytes(' [...]\n', encoding='latin1') _COMMENT_START = as_bytes('<!-- ', encoding='latin1') _COMMENT_END = as_bytes(' -->', encoding='latin1') _PI_START = as_bytes('<?', encoding='latin1') _PI_END = as_bytes('?>', encoding='latin1') _AMPER_ESCAPED = as_bytes('&', encoding='latin1') _LT = as_bytes('<', encoding='latin1') _LT_ESCAPED = as_bytes('<', encoding='latin1') _QUOTE_ESCAPED = as_bytes(""", encoding='latin1') _XML_PROLOG_BEGIN = as_bytes('<?xml version="1.0"', encoding='latin1') _ENCODING = as_bytes('encoding', encoding='latin1') _XML_PROLOG_END = as_bytes('?>\n', encoding='latin1') _DOCTYPE_BEGIN = as_bytes('<!DOCTYPE', encoding='latin1') _PUBLIC = as_bytes('PUBLIC', encoding='latin1') _DOCTYPE_END = as_bytes('>\n', encoding='latin1') if PY2: def encode(text, encoding): return text.encode(encoding) else: def encode(text, encoding): if not isinstance(text, bytes): text = text.encode(encoding) return text # replace element factory def Replace(text, structure=False): element = _MeldElementInterface(Replace, {}) element.text = text element.structure = structure return element class PyHelper: def findmeld(self, node, name, default=None): iterator = self.getiterator(node) for element in iterator: val = element.attrib.get(_MELD_ID) if val == name: return element return default def clone(self, node, parent=None): element = _MeldElementInterface(node.tag, node.attrib.copy()) element.text = node.text element.tail = node.tail element.structure = node.structure if parent is not None: # avoid calling self.append to reduce function call overhead parent._children.append(element) element.parent = parent for child in node._children: self.clone(child, element) return element def _bfclone(self, nodes, parent): L = [] for node in nodes: element = _MeldElementInterface(node.tag, node.attrib.copy()) element.parent = parent element.text = node.text element.tail = node.tail element.structure = node.structure if node._children: self._bfclone(node._children, element) L.append(element) parent._children = L def bfclone(self, node, parent=None): element = _MeldElementInterface(node.tag, node.attrib.copy()) element.text = node.text element.tail = node.tail element.structure = node.structure element.parent = parent if parent is not None: parent._children.append(element) if node._children: self._bfclone(node._children, element) return element def getiterator(self, node, tag=None): nodes = [] if tag == "*": tag = None if tag is None or node.tag == tag: nodes.append(node) for element in node._children: nodes.extend(self.getiterator(element, tag)) return nodes def content(self, node, text, structure=False): node.text = None replacenode = Replace(text, structure) replacenode.parent = node replacenode.text = text replacenode.structure = structure node._children = [replacenode] helper = PyHelper() _MELD_NS_URL = 'https://github.com/Supervisor/supervisor' _MELD_PREFIX = '{%s}' % _MELD_NS_URL _MELD_LOCAL = 'id' _MELD_ID = '%s%s' % (_MELD_PREFIX, _MELD_LOCAL) _MELD_SHORT_ID = 'meld:%s' % _MELD_LOCAL _XHTML_NS_URL = 'http://www.w3.org/1999/xhtml' _XHTML_PREFIX = '{%s}' % _XHTML_NS_URL _XHTML_PREFIX_LEN = len(_XHTML_PREFIX) _marker = [] class doctype: # lookup table for ease of use in external code html_strict = ('HTML', '-//W3C//DTD HTML 4.01//EN', 'http://www.w3.org/TR/html4/strict.dtd') html = ('HTML', '-//W3C//DTD HTML 4.01 Transitional//EN', 'http://www.w3.org/TR/html4/loose.dtd') xhtml_strict = ('html', '-//W3C//DTD XHTML 1.0 Strict//EN', 'http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd') xhtml = ('html', '-//W3C//DTD XHTML 1.0 Transitional//EN', 'http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd') class _MeldElementInterface: parent = None attrib = None text = None tail = None structure = None # overrides to reduce MRU lookups def __init__(self, tag, attrib): self.tag = tag self.attrib = attrib self._children = [] def __repr__(self): return "<MeldElement %s at %x>" % (self.tag, id(self)) def __len__(self): return len(self._children) def __getitem__(self, index): return self._children[index] def __getslice__(self, start, stop): return self._children[start:stop] def getchildren(self): return self._children def find(self, path): return ElementPath.find(self, path) def findtext(self, path, default=None): return ElementPath.findtext(self, path, default) def findall(self, path): return ElementPath.findall(self, path) def clear(self): self.attrib.clear() self._children = [] self.text = self.tail = None def get(self, key, default=None): return self.attrib.get(key, default) def set(self, key, value): self.attrib[key] = value def keys(self): return list(self.attrib.keys()) def items(self): return list(self.attrib.items()) def getiterator(self, *ignored_args, **ignored_kw): # we ignore any tag= passed in to us, originally because it was too # painfail to support in the old C extension, now for b/w compat return helper.getiterator(self) # overrides to support parent pointers and factories def __setitem__(self, index, element): if isinstance(index, slice): for e in element: e.parent = self else: element.parent = self self._children[index] = element # TODO: Can __setslice__ be removed now? def __setslice__(self, start, stop, elements): for element in elements: element.parent = self self._children[start:stop] = list(elements) def append(self, element): self._children.append(element) element.parent = self def insert(self, index, element): self._children.insert(index, element) element.parent = self def __delitem__(self, index): if isinstance(index, slice): for ob in self._children[index]: ob.parent = None else: self._children[index].parent = None ob = self._children[index] del self._children[index] # TODO: Can __delslice__ be removed now? def __delslice__(self, start, stop): obs = self._children[start:stop] for ob in obs: ob.parent = None del self._children[start:stop] def remove(self, element): self._children.remove(element) element.parent = None def makeelement(self, tag, attrib): return self.__class__(tag, attrib) # meld-specific def __mod__(self, other): """ Fill in the text values of meld nodes in tree; only support dictionarylike operand (sequence operand doesn't seem to make sense here)""" return self.fillmelds(**other) def fillmelds(self, **kw): """ Fill in the text values of meld nodes in tree using the keyword arguments passed in; use the keyword keys as meld ids and the keyword values as text that should fill in the node text on which that meld id is found. Return a list of keys from **kw that were not able to be found anywhere in the tree. Never raises an exception. """ unfilled = [] for k in kw: node = self.findmeld(k) if node is None: unfilled.append(k) else: node.text = kw[k] return unfilled def fillmeldhtmlform(self, **kw): """ Perform magic to 'fill in' HTML form element values from a dictionary. Unlike 'fillmelds', the type of element being 'filled' is taken into consideration. Perform a 'findmeld' on each key in the dictionary and use the value that corresponds to the key to perform mutation of the tree, changing data in what is presumed to be one or more HTML form elements according to the following rules:: If the found element is an 'input group' (its meld id ends with the string ':inputgroup'), set the 'checked' attribute on the appropriate subelement which has a 'value' attribute which matches the dictionary value. Also remove the 'checked' attribute from every other 'input' subelement of the input group. If no input subelement's value matches the dictionary value, this key is treated as 'unfilled'. If the found element is an 'input type=text', 'input type=hidden', 'input type=submit', 'input type=password', 'input type=reset' or 'input type=file' element, replace its 'value' attribute with the value. If the found element is an 'input type=checkbox' or 'input type='radio' element, set its 'checked' attribute to true if the dict value is true, or remove its 'checked' attribute if the dict value is false. If the found element is a 'select' element and the value exists in the 'value=' attribute of one of its 'option' subelements, change that option's 'selected' attribute to true and mark all other option elements as unselected. If the select element does not contain an option with a value that matches the dictionary value, do nothing and return this key as unfilled. If the found element is a 'textarea' or any other kind of element, replace its text with the value. If the element corresponding to the key is not found, do nothing and treat the key as 'unfilled'. Return a list of 'unfilled' keys, representing meld ids present in the dictionary but not present in the element tree or meld ids which could not be filled due to the lack of any matching subelements for 'select' nodes or 'inputgroup' nodes. """ unfilled = [] for k in kw: node = self.findmeld(k) if node is None: unfilled.append(k) continue val = kw[k] if k.endswith(':inputgroup'): # an input group is a list of input type="checkbox" or # input type="radio" elements that can be treated as a group # because they attempt to specify the same value found = [] unfound = [] for child in node.findall('input'): input_type = child.attrib.get('type', '').lower() if input_type not in ('checkbox', 'radio'): continue input_val = child.attrib.get('value', '') if val == input_val: found.append(child) else: unfound.append(child) if not found: unfilled.append(k) else: for option in found: option.attrib['checked'] = 'checked' for option in unfound: try: del option.attrib['checked'] except KeyError: pass else: tag = node.tag.lower() if tag == 'input': input_type = node.attrib.get('type', 'text').lower() # fill in value attrib for most input types if input_type in ('hidden', 'submit', 'text', 'password', 'reset', 'file'): node.attrib['value'] = val # unless it's a checkbox or radio attribute, then we # fill in its checked attribute elif input_type in ('checkbox', 'radio'): if val: node.attrib['checked'] = 'checked' else: try: del node.attrib['checked'] except KeyError: pass else: unfilled.append(k) elif tag == 'select': # if the node is a select node, we want to select # the value matching val, otherwise it's unfilled found = [] unfound = [] for option in node.findall('option'): if option.attrib.get('value', '') == val: found.append(option) else: unfound.append(option) if not found: unfilled.append(k) else: for option in found: option.attrib['selected'] = 'selected' for option in unfound: try: del option.attrib['selected'] except KeyError: pass else: node.text = kw[k] return unfilled def findmeld(self, name, default=None): """ Find a node in the tree that has a 'meld id' corresponding to 'name'. Iterate over all subnodes recursively looking for a node which matches. If we can't find the node, return None.""" # this could be faster if we indexed all the meld nodes in the # tree; we just walk the whole hierarchy now. result = helper.findmeld(self, name) if result is None: return default return result def findmelds(self): """ Find all nodes that have a meld id attribute and return the found nodes in a list""" return self.findwithattrib(_MELD_ID) def findwithattrib(self, attrib, value=None): """ Find all nodes that have an attribute named 'attrib'. If 'value' is not None, omit nodes on which the attribute value does not compare equally to 'value'. Return the found nodes in a list.""" iterator = helper.getiterator(self) elements = [] for element in iterator: attribval = element.attrib.get(attrib) if attribval is not None: if value is None: elements.append(element) else: if value == attribval: elements.append(element) return elements # ZPT-alike methods def repeat(self, iterable, childname=None): """repeats an element with values from an iterable. If 'childname' is not None, repeat the element on which the repeat is called, otherwise find the child element with a 'meld:id' matching 'childname' and repeat that. The element is repeated within its parent element (nodes that are created as a result of a repeat share the same parent). This method returns an iterable; the value of each iteration is a two-sequence in the form (newelement, data). 'newelement' is a clone of the template element (including clones of its children) which has already been seated in its parent element in the template. 'data' is a value from the passed in iterable. Changing 'newelement' (typically based on values from 'data') mutates the element 'in place'.""" if childname: element = self.findmeld(childname) else: element = self parent = element.parent # creating a list is faster than yielding a generator (py 2.4) L = [] first = True for thing in iterable: if first is True: clone = element else: clone = helper.bfclone(element, parent) L.append((clone, thing)) first = False return L def replace(self, text, structure=False): """ Replace this element with a Replace node in our parent with the text 'text' and return the index of our position in our parent. If we have no parent, do nothing, and return None. Pass the 'structure' flag to the replace node so it can do the right thing at render time. """ parent = self.parent i = self.deparent() if i is not None: # reduce function call overhead by not calling self.insert node = Replace(text, structure) parent._children.insert(i, node) node.parent = parent return i def content(self, text, structure=False): """ Delete this node's children and append a Replace node that contains text. Always return None. Pass the 'structure' flag to the replace node so it can do the right thing at render time.""" helper.content(self, text, structure) def attributes(self, **kw): """ Set attributes on this node. """ for k, v in kw.items(): # prevent this from getting to the parser if possible if not isinstance(k, StringTypes): raise ValueError('do not set non-stringtype as key: %s' % k) if not isinstance(v, StringTypes): raise ValueError('do not set non-stringtype as val: %s' % v) self.attrib[k] = kw[k] # output methods def write_xmlstring(self, encoding=None, doctype=None, fragment=False, declaration=True, pipeline=False): data = [] write = data.append if not fragment: if declaration: _write_declaration(write, encoding) if doctype: _write_doctype(write, doctype) _write_xml(write, self, encoding, {}, pipeline) return _BLANK.join(data) def write_xml(self, file, encoding=None, doctype=None, fragment=False, declaration=True, pipeline=False): """ Write XML to 'file' (which can be a filename or filelike object) encoding - encoding string (if None, 'utf-8' encoding is assumed) Must be a recognizable Python encoding type. doctype - 3-tuple indicating name, pubid, system of doctype. The default is to prevent a doctype from being emitted. fragment - True if a 'fragment' should be emitted for this node (no declaration, no doctype). This causes both the 'declaration' and 'doctype' parameters to become ignored if provided. declaration - emit an xml declaration header (including an encoding if it's not None). The default is to emit the doctype. pipeline - preserve 'meld' namespace identifiers in output for use in pipelining """ if not hasattr(file, "write"): file = open(file, "wb") data = self.write_xmlstring(encoding, doctype, fragment, declaration, pipeline) file.write(data) def write_htmlstring(self, encoding=None, doctype=doctype.html, fragment=False): data = [] write = data.append if encoding is None: encoding = 'utf8' if not fragment: if doctype: _write_doctype(write, doctype) _write_html(write, self, encoding, {}) joined = _BLANK.join(data) return joined def write_html(self, file, encoding=None, doctype=doctype.html, fragment=False): """ Write HTML to 'file' (which can be a filename or filelike object) encoding - encoding string (if None, 'utf-8' encoding is assumed). Unlike XML output, this is not used in a declaration, but it is used to do actual character encoding during output. Must be a recognizable Python encoding type. doctype - 3-tuple indicating name, pubid, system of doctype. The default is the value of doctype.html (HTML 4.0 'loose') fragment - True if a "fragment" should be omitted (no doctype). This overrides any provided "doctype" parameter if provided. Namespace'd elements and attributes have their namespaces removed during output when writing HTML, so pipelining cannot be performed. HTML is not valid XML, so an XML declaration header is never emitted. """ if not hasattr(file, "write"): file = open(file, "wb") page = self.write_htmlstring(encoding, doctype, fragment) file.write(page) def write_xhtmlstring(self, encoding=None, doctype=doctype.xhtml, fragment=False, declaration=False, pipeline=False): data = [] write = data.append if not fragment: if declaration: _write_declaration(write, encoding) if doctype: _write_doctype(write, doctype) _write_xml(write, self, encoding, {}, pipeline, xhtml=True) return _BLANK.join(data) def write_xhtml(self, file, encoding=None, doctype=doctype.xhtml, fragment=False, declaration=False, pipeline=False): """ Write XHTML to 'file' (which can be a filename or filelike object) encoding - encoding string (if None, 'utf-8' encoding is assumed) Must be a recognizable Python encoding type. doctype - 3-tuple indicating name, pubid, system of doctype. The default is the value of doctype.xhtml (XHTML 'loose'). fragment - True if a 'fragment' should be emitted for this node (no declaration, no doctype). This causes both the 'declaration' and 'doctype' parameters to be ignored. declaration - emit an xml declaration header (including an encoding string if 'encoding' is not None) pipeline - preserve 'meld' namespace identifiers in output for use in pipelining """ if not hasattr(file, "write"): file = open(file, "wb") page = self.write_xhtmlstring(encoding, doctype, fragment, declaration, pipeline) file.write(page) def clone(self, parent=None): """ Create a clone of an element. If parent is not None, append the element to the parent. Recurse as necessary to create a deep clone of the element. """ return helper.bfclone(self, parent) def deparent(self): """ Remove ourselves from our parent node (de-parent) and return the index of the parent which was deleted. """ i = self.parentindex() if i is not None: del self.parent[i] return i def parentindex(self): """ Return the parent node index in which we live """ parent = self.parent if parent is not None: return parent._children.index(self) def shortrepr(self, encoding=None): data = [] _write_html(data.append, self, encoding, {}, maxdepth=2) return _BLANK.join(data) def diffmeld(self, other): """ Compute the meld element differences from this node (the source) to 'other' (the target). Return a dictionary of sequences in the form {'unreduced: {'added':[], 'removed':[], 'moved':[]}, 'reduced': {'added':[], 'removed':[], 'moved':[]},} """ srcelements = self.findmelds() tgtelements = other.findmelds() srcids = [ x.meldid() for x in srcelements ] tgtids = [ x.meldid() for x in tgtelements ] removed = [] for srcelement in srcelements: if srcelement.meldid() not in tgtids: removed.append(srcelement) added = [] for tgtelement in tgtelements: if tgtelement.meldid() not in srcids: added.append(tgtelement) moved = [] for srcelement in srcelements: srcid = srcelement.meldid() if srcid in tgtids: i = tgtids.index(srcid) tgtelement = tgtelements[i] if not sharedlineage(srcelement, tgtelement): moved.append(tgtelement) unreduced = {'added':added, 'removed':removed, 'moved':moved} moved_reduced = diffreduce(moved) added_reduced = diffreduce(added) removed_reduced = diffreduce(removed) reduced = {'moved':moved_reduced, 'added':added_reduced, 'removed':removed_reduced} return {'unreduced':unreduced, 'reduced':reduced} def meldid(self): return self.attrib.get(_MELD_ID) def lineage(self): L = [] parent = self while parent is not None: L.append(parent) parent = parent.parent return L class MeldTreeBuilder(TreeBuilder): def __init__(self): TreeBuilder.__init__(self, element_factory=_MeldElementInterface) self.meldids = {} def start(self, tag, attrs): elem = TreeBuilder.start(self, tag, attrs) for key, value in attrs.items(): if key == _MELD_ID: if value in self.meldids: raise ValueError('Repeated meld id "%s" in source' % value) self.meldids[value] = 1 break return elem def comment(self, data): self.start(Comment, {}) self.data(data) self.end(Comment) def doctype(self, name, pubid, system): pass class HTMLXMLParser(HTMLParser): """ A mostly-cut-and-paste of ElementTree's HTMLTreeBuilder that does special meld3 things (like preserve comments and munge meld ids). Subclassing is not possible due to private attributes. :-(""" def __init__(self, builder=None, encoding=None): self.__stack = [] if builder is None: builder = MeldTreeBuilder() self.builder = builder self.encoding = encoding or "iso-8859-1" try: # ``convert_charrefs`` was added in Python 3.4. Set it to avoid # "DeprecationWarning: The value of convert_charrefs will become # True in 3.5. You are encouraged to set the value explicitly." HTMLParser.__init__(self, convert_charrefs=False) except TypeError: HTMLParser.__init__(self) self.meldids = {} def close(self): HTMLParser.close(self) self.meldids = {} return self.builder.close() def handle_starttag(self, tag, attrs): if tag == "meta": # look for encoding directives http_equiv = content = None for k, v in attrs: if k == "http-equiv": http_equiv = v.lower() elif k == "content": content = v if http_equiv == "content-type" and content: # use email to parse the http header msg = email.message_from_string( "%s: %s\n\n" % (http_equiv, content) ) encoding = msg.get_param("charset") if encoding: self.encoding = encoding if tag in AUTOCLOSE: if self.__stack and self.__stack[-1] == tag: self.handle_endtag(tag) self.__stack.append(tag) attrib = {} if attrs: for k, v in attrs: if k == _MELD_SHORT_ID: k = _MELD_ID if self.meldids.get(v): raise ValueError('Repeated meld id "%s" in source' % v) self.meldids[v] = 1 else: k = k.lower() attrib[k] = v self.builder.start(tag, attrib) if tag in IGNOREEND: self.__stack.pop() self.builder.end(tag) def handle_endtag(self, tag): if tag in IGNOREEND: return lasttag = self.__stack.pop() if tag != lasttag and lasttag in AUTOCLOSE: self.handle_endtag(lasttag) self.builder.end(tag) def handle_charref(self, char): if char[:1] == "x": char = int(char[1:], 16) else: char = int(char) self.builder.data(unichr(char)) def handle_entityref(self, name): entity = htmlentitydefs.entitydefs.get(name) if entity: if len(entity) == 1: entity = ord(entity) else: entity = int(entity[2:-1]) self.builder.data(unichr(entity)) else: self.unknown_entityref(name) def handle_data(self, data): if isinstance(data, bytes): data = as_string(data, self.encoding) self.builder.data(data) def unknown_entityref(self, name): pass # ignore by default; override if necessary def handle_comment(self, data): self.builder.start(Comment, {}) self.builder.data(data) self.builder.end(Comment) def do_parse(source, parser): root = et_parse(source, parser=parser).getroot() iterator = root.getiterator() for p in iterator: for c in p: c.parent = p return root def parse_xml(source): """ Parse source (a filelike object) into an element tree. If html is true, use a parser that can resolve somewhat ambiguous HTML into XHTML. Otherwise use a 'normal' parser only.""" builder = MeldTreeBuilder() parser = XMLParser(target=builder) return do_parse(source, parser) def parse_html(source, encoding=None): builder = MeldTreeBuilder() parser = HTMLXMLParser(builder, encoding) return do_parse(source, parser) def parse_xmlstring(text): source = StringIO(text) return parse_xml(source) def parse_htmlstring(text, encoding=None): source = StringIO(text) return parse_html(source, encoding) attrib_needs_escaping = re.compile(r'[&"<]').search cdata_needs_escaping = re.compile(r'[&<]').search def _both_case(mapping): # Add equivalent upper-case keys to mapping. lc_keys = list(mapping.keys()) for k in lc_keys: mapping[k.upper()] = mapping[k] _HTMLTAGS_UNBALANCED = {'area':1, 'base':1, 'basefont':1, 'br':1, 'col':1, 'frame':1, 'hr':1, 'img':1, 'input':1, 'isindex':1, 'link':1, 'meta':1, 'param':1} _both_case(_HTMLTAGS_UNBALANCED) _HTMLTAGS_NOESCAPE = {'script':1, 'style':1} _both_case(_HTMLTAGS_NOESCAPE) _HTMLATTRS_BOOLEAN = {'selected':1, 'checked':1, 'compact':1, 'declare':1, 'defer':1, 'disabled':1, 'ismap':1, 'multiple':1, 'nohref':1, 'noresize':1, 'noshade':1, 'nowrap':1} _both_case(_HTMLATTRS_BOOLEAN) def _write_html(write, node, encoding, namespaces, depth=-1, maxdepth=None): """ Walk 'node', calling 'write' with bytes(?). """ if encoding is None: encoding = 'utf-8' tag = node.tag tail = node.tail text = node.text tail = node.tail to_write = _BLANK if tag is Replace: if not node.structure: if cdata_needs_escaping(text): text = _escape_cdata(text) write(encode(text, encoding)) elif tag is Comment: if cdata_needs_escaping(text): text = _escape_cdata(text) write(encode('<!-- ' + text + ' -->', encoding)) elif tag is ProcessingInstruction: if cdata_needs_escaping(text): text = _escape_cdata(text) write(encode('<!-- ' + text + ' -->', encoding)) else: xmlns_items = [] # new namespaces in this scope try: if tag[:1] == "{": if tag[:_XHTML_PREFIX_LEN] == _XHTML_PREFIX: tag = tag[_XHTML_PREFIX_LEN:] else: tag, xmlns = fixtag(tag, namespaces) if xmlns: xmlns_items.append(xmlns) except TypeError: _raise_serialization_error(tag) to_write += _OPEN_TAG_START + encode(tag, encoding) attrib = node.attrib if attrib is not None: if len(attrib) > 1: attrib_keys = list(attrib.keys()) attrib_keys.sort() else: attrib_keys = attrib for k in attrib_keys: try: if k[:1] == "{": continue except TypeError: _raise_serialization_error(k) if k in _HTMLATTRS_BOOLEAN: to_write += _SPACE + encode(k, encoding) else: v = attrib[k] to_write += _encode_attrib(k, v, encoding) for k, v in xmlns_items: to_write += _encode_attrib(k, v, encoding) to_write += _OPEN_TAG_END if text is not None and text: if tag in _HTMLTAGS_NOESCAPE: to_write += encode(text, encoding) elif cdata_needs_escaping(text): to_write += _escape_cdata(text) else: to_write += encode(text,encoding) write(to_write) for child in node._children: if maxdepth is not None: depth = depth + 1 if depth < maxdepth: _write_html(write, child, encoding, namespaces, depth, maxdepth) elif depth == maxdepth and text: write(_OMITTED_TEXT) else: _write_html(write, child, encoding, namespaces, depth, maxdepth) if text or node._children or tag not in _HTMLTAGS_UNBALANCED: write(_CLOSE_TAG_START + encode(tag, encoding) + _CLOSE_TAG_END) if tail: if cdata_needs_escaping(tail): write(_escape_cdata(tail)) else: write(encode(tail,encoding)) def _write_xml(write, node, encoding, namespaces, pipeline, xhtml=False): """ Write XML to a file """ if encoding is None: encoding = 'utf-8' tag = node.tag if tag is Comment: write(_COMMENT_START + _escape_cdata(node.text, encoding) + _COMMENT_END) elif tag is ProcessingInstruction: write(_PI_START + _escape_cdata(node.text, encoding) + _PI_END) elif tag is Replace: if node.structure: # this may produce invalid xml write(encode(node.text, encoding)) else: write(_escape_cdata(node.text, encoding)) else: if xhtml: if tag[:_XHTML_PREFIX_LEN] == _XHTML_PREFIX: tag = tag[_XHTML_PREFIX_LEN:] if node.attrib: items = list(node.attrib.items()) else: items = [] # must always be sortable. xmlns_items = [] # new namespaces in this scope try: if tag[:1] == "{": tag, xmlns = fixtag(tag, namespaces) if xmlns: xmlns_items.append(xmlns) except TypeError: _raise_serialization_error(tag) write(_OPEN_TAG_START + encode(tag, encoding)) if items or xmlns_items: items.sort() # lexical order for k, v in items: try: if k[:1] == "{": if not pipeline: if k == _MELD_ID: continue k, xmlns = fixtag(k, namespaces) if xmlns: xmlns_items.append(xmlns) if not pipeline: # special-case for HTML input if k == 'xmlns:meld': continue except TypeError: _raise_serialization_error(k) write(_encode_attrib(k, v, encoding)) for k, v in xmlns_items: write(_encode_attrib(k, v, encoding)) if node.text or node._children: write(_OPEN_TAG_END) if node.text: write(_escape_cdata(node.text, encoding)) for n in node._children: _write_xml(write, n, encoding, namespaces, pipeline, xhtml) write(_CLOSE_TAG_START + encode(tag, encoding) + _CLOSE_TAG_END) else: write(_SELF_CLOSE) for k, v in xmlns_items: del namespaces[v] if node.tail: write(_escape_cdata(node.tail, encoding)) def _encode_attrib(k, v, encoding): return _BLANK.join((_SPACE, encode(k, encoding), _EQUAL, _QUOTE, _escape_attrib(v, encoding), _QUOTE, )) # overrides to elementtree to increase speed and get entity quoting correct. # negative lookahead assertion _NONENTITY_RE = re.compile(as_bytes(r'&(?!([#\w]*;))', encoding='latin1')) def _escape_cdata(text, encoding=None): # Return escaped character data as bytes. try: if encoding: try: encoded = encode(text, encoding) except UnicodeError: return _encode_entity(text) else: encoded = as_bytes(text, encoding='latin1') encoded = _NONENTITY_RE.sub(_AMPER_ESCAPED, encoded) encoded = encoded.replace(_LT, _LT_ESCAPED) return encoded except (TypeError, AttributeError): _raise_serialization_error(text) def _escape_attrib(text, encoding): # Return escaped attribute value as bytes. try: if encoding: try: encoded = encode(text, encoding) except UnicodeError: return _encode_entity(text) else: encoded = as_bytes(text, encoding='latin1') # don't requote properly-quoted entities encoded = _NONENTITY_RE.sub(_AMPER_ESCAPED, encoded) encoded = encoded.replace(_LT, _LT_ESCAPED) encoded = encoded.replace(_QUOTE, _QUOTE_ESCAPED) return encoded except (TypeError, AttributeError): _raise_serialization_error(text) # utility functions def _write_declaration(write, encoding): # Write as bytes. if not encoding: write(_XML_PROLOG_BEGIN + _XML_PROLOG_END) else: write(_XML_PROLOG_BEGIN + _SPACE + _ENCODING + _EQUAL + _QUOTE + as_bytes(encoding, encoding='latin1') + _QUOTE + _XML_PROLOG_END) def _write_doctype(write, doctype): # Write as bytes. try: name, pubid, system = doctype except (ValueError, TypeError): raise ValueError("doctype must be supplied as a 3-tuple in the form " "(name, pubid, system) e.g. '%s'" % doctype.xhtml) write(_DOCTYPE_BEGIN + _SPACE + as_bytes(name, encoding='latin1') + _SPACE + _PUBLIC + _SPACE + _QUOTE + as_bytes(pubid, encoding='latin1') + _QUOTE + _SPACE + _QUOTE + as_bytes(system, encoding='latin1') + _QUOTE + _DOCTYPE_END) _XML_DECL_RE = re.compile(r'<\?xml .*?\?>') _BEGIN_TAG_RE = re.compile(r'<[^/?!]?\w+') def insert_doctype(data, doctype=doctype.xhtml): # jam an html doctype declaration into 'data' if it # doesn't already contain a doctype declaration match = _XML_DECL_RE.search(data) dt_string = '<!DOCTYPE %s PUBLIC "%s" "%s">' % doctype if match is not None: start, end = match.span(0) before = data[:start] tag = data[start:end] after = data[end:] return before + tag + dt_string + after else: return dt_string + data def insert_meld_ns_decl(data): match = _BEGIN_TAG_RE.search(data) if match is not None: start, end = match.span(0) before = data[:start] tag = data[start:end] + ' xmlns:meld="%s"' % _MELD_NS_URL after = data[end:] data = before + tag + after return data def prefeed(data, doctype=doctype.xhtml): if data.find('<!DOCTYPE') == -1: data = insert_doctype(data, doctype) if data.find('xmlns:meld') == -1: data = insert_meld_ns_decl(data) return data def sharedlineage(srcelement, tgtelement): srcparent = srcelement.parent tgtparent = tgtelement.parent srcparenttag = getattr(srcparent, 'tag', None) tgtparenttag = getattr(tgtparent, 'tag', None) if srcparenttag != tgtparenttag: return False elif tgtparenttag is None and srcparenttag is None: return True elif tgtparent and srcparent: return sharedlineage(srcparent, tgtparent) return False def diffreduce(elements): # each element in 'elements' should all have non-None meldids, and should # be preordered in depth-first traversal order reduced = [] for element in elements: parent = element.parent if parent is None: reduced.append(element) continue if parent in reduced: continue reduced.append(element) return reduced def intersection(S1, S2): L = [] for element in S1: if element in S2: L.append(element) return L def melditerator(element, meldid=None, _MELD_ID=_MELD_ID): nodeid = element.attrib.get(_MELD_ID) if nodeid is not None: if meldid is None or nodeid == meldid: yield element for child in element._children: for el2 in melditerator(child, meldid): nodeid = el2.attrib.get(_MELD_ID) if nodeid is not None: if meldid is None or nodeid == meldid: yield el2 #----------------------------------------------------------------------------- # Begin fork from Python 2.6.8 stdlib: # - xml.elementtree.ElementTree._raise_serialization_error # - xml.elementtree.ElementTree._encode_entity # - xml.elementtree.ElementTree._namespace_map # - xml.elementtree.ElementTree.fixtag #----------------------------------------------------------------------------- _NON_ASCII_MIN = as_string('\xc2\x80', 'utf-8') # u'\u0080' _NON_ASCII_MAX = as_string('\xef\xbf\xbf', 'utf-8') # u'\uffff' _escape_map = { "&": "&", "<": "<", ">": ">", '"': """, } _namespace_map = { # "well-known" namespace prefixes "http://www.w3.org/XML/1998/namespace": "xml", "http://www.w3.org/1999/xhtml": "html", "http://www.w3.org/1999/02/22-rdf-syntax-ns#": "rdf", "http://schemas.xmlsoap.org/wsdl/": "wsdl", } def _encode(s, encoding): try: return s.encode(encoding) except AttributeError: return s def _raise_serialization_error(text): raise TypeError( "cannot serialize %r (type %s)" % (text, type(text).__name__) ) _pattern = None def _encode_entity(text): # map reserved and non-ascii characters to numerical entities global _pattern if _pattern is None: _ptxt = r'[&<>\"' + _NON_ASCII_MIN + '-' + _NON_ASCII_MAX + ']+' #_pattern = re.compile(eval(r'u"[&<>\"\u0080-\uffff]+"')) _pattern = re.compile(_ptxt) def _escape_entities(m): out = [] append = out.append for char in m.group(): text = _escape_map.get(char) if text is None: text = "&#%d;" % ord(char) append(text) return ''.join(out) try: return _encode(_pattern.sub(_escape_entities, text), "ascii") except TypeError: _raise_serialization_error(text) def fixtag(tag, namespaces): # given a decorated tag (of the form {uri}tag), return prefixed # tag and namespace declaration, if any if isinstance(tag, QName): tag = tag.text namespace_uri, tag = tag[1:].split("}", 1) prefix = namespaces.get(namespace_uri) if prefix is None: prefix = _namespace_map.get(namespace_uri) if prefix is None: prefix = "ns%d" % len(namespaces) namespaces[namespace_uri] = prefix if prefix == "xml": xmlns = None else: xmlns = ("xmlns:%s" % prefix, namespace_uri) else: xmlns = None return "%s:%s" % (prefix, tag), xmlns #----------------------------------------------------------------------------- # End fork from Python 2.6.8 stdlib #-----------------------------------------------------------------------------